# AGI House case — provenance

Generated 2026-08-04 by lane BUILD-A. Source tree treated as read-only.

## Screens

| Screen | Source file lifted | Changes | Anonymisation |
|--------|-------------------|---------|---------------|
| Case wrapper | *(new)* `06_cases/agi-house/index.html` | Portfolio narrative only | N/A — no personal data |
| People explorer | `public/agi-house/people.html` + inline JS + `style.css` | Data URLs → `../data/`; removed Google Fonts / CDN emoji / remote logo img; absolute applicants candidates neutralized | Working-set `applicants.json` (312) fully rewritten: deterministic name aliases, `person+NNNN@example.test`, employers → sector·size (public sponsor/uni brands kept), LinkedIn/social/avatar URLs dropped, score **numbers** kept, free-text comments replaced with numeric summaries |
| Firecrawl event report | `public/agi-house/report-event-self-evolving-agent-build-day-20251101.html` | Same path/offline patches; people iframe → local `people.html` | Uses anonymised applicants/events/projects; report aggregates from `agi_house_report.json` with email lists replaced; public sponsor brands retained (incl. Firecrawl) |
| Impact people / companies | `public/agi-house/impact/index.html` + `agi-h-impact.js` + `agi-h-impact.css` + `events-index.json` | Removed GitHub Pages v2 remote override; `__AGIH_IMPACT_DATA_URL__` / `__AGIH_IMPACT_V2_URL__` → anonymised v1 fixture (v2 file absent in source) | `founders-impact.v1.json` (165): synthetic person/company names, provenance methods kept, estimation HTML scrubbed, dates month-precision, elevation **numbers** retained |
| Event shell | `public/agi-house/index.html` + `script.js` | Data URLs → `../data/` | Same events/applicants fixtures |
| Companies | `public/agi-house/companies.html` | Data URL → anonymised `companies.json` | Company names → sector aliases; URLs/sources cleared; group/tags kept |
| Events | `public/agi-house/events.html` | Served via `script.js` / catalog | Speakers/judges/organizers aliased; public sponsors kept; media URLs cleared |

## Data files in `data/`

| File | Source | Transform |
|------|--------|-----------|
| `applicants.json` | `public/agi-house/applicants.json` | Anonymise (see above). Counts preserved: 312 / scores_v2 200 / step 8.1+8.2 = 202 |
| `events.json` | `events.json` | Anonymise people chips; keep event ids/titles; sponsors public brands |
| `companies.json` | `companies.json` | Synthetic company labels; 151 rows |
| `projects.json` | `projects.json` | Member names aliased; URLs cleared; 25 rows |
| `founders-impact.v1.json` | `impact/founders-impact.v1.json` | Synthetic identities; 165 rows / 122 company ids (count from source) |
| `agi_house_report.json` | `agi_house_report.json` | Aggregates kept (`total_with_company_data: 250`); people emails synthetic |
| `AGI_House_Judge_Sheet__11_1_1.json` | same | Row names/emails replaced; column headers that embedded judge first names renamed to `Judge N: JudgeAlias`; project titles hashed |
| `golden-tags.json`, `tag-catalog.json`, `stats.json`, `dashboard-config.json` | same | Copied (taxonomy / empty stats — no PII) |
| `candidates-outreach.json` | same | Emptied (`[]`) — outreach list omitted |
| `v2.additional_applicants_base` | same | Empty stub |

## Explicitly not shipped

- `Full-enriched-DB.json` (13,733 objects) — count cited only
- `impact/Final_2023-hack-DB.json` (2,391) — count cited only
- `dinner-guests.json`, `impact/all_dinner_guests-analysis.json` — omitted (PII / corrupt JSON)
- `New_founders-impact.v2.json` — absent in source; not invented
- Any `.env`, credentials, or MCP configs — none present in the lifted paths

## Secrets / telemetry removed

- No `.env` or credential files encountered in the lifted set.
- Remote GitHub raw v2 URL override in impact index neutralized.
- External font/CDN/logo image dependencies removed or pointed at local avatar SVG so the case runs offline for **data**; outbound sponsor homepage links may remain as plain anchors.

## Tooling

- `tools/anonymize_agi.py` — regenerates `app/` + `data/` from the read-only source tree.


## Post-build scrub (2026-08-04)
- Regenerated data via tools/anonymize_agi.py after emails were corrupted to a literal placeholder.
- Removed owner-specific score override from app/people.html.
- Renamed event id containing a personal name to nanda-mcp-hack-20250423.
- post_lift_scrub now runs at end of anonymiser.